In this second notebook, we use the cleaned and filtered data produced from notebook 1. Here we explore the spatial distribution of our dataset.
So far, we worked with pandas DataFrame objects, to map the properties, we import the ArcGIS API for Python. This adds a GeoAccessor and spatially enables your DataFrame objects. As you will see in this notebook, you can then easily plot your data both as charts and as maps.
import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
from arcgis.gis import GIS
from arcgis.features import GeoAccessor, GeoSeriesAccessor
gis = GIS()
csv_path = 'resources/houses_for_sale_filtered.csv'
prop_df = pd.read_csv(csv_path)
prop_df.head(3)
prop_df.shape
prop_df.columns
Drop redundant columns
try:
prop_df.drop(columns=['Unnamed: 0'], inplace=True)
prop_df.drop(columns=['Unnamed: 0.1'], inplace=True)
except:
pass
Convert to Spatially Enabled DataFrame
prop_sdf = pd.DataFrame.spatial.from_xy(prop_df, 'LONGITUDE','LATITUDE')
type(prop_sdf)
pdx_map = gis.map('Portland, OR')
pdx_map.basemap = 'streets'
pdx_map

prop_sdf.spatial.plot(map_widget=pdx_map)
You can render the same data using a heatmap renderer, which visualizes the spatial density of the properties (houses) using a heatmap.
pdx_density_map = gis.map('Portland, OR')
pdx_density_map.basemap='gray'
pdx_density_map

prop_sdf.spatial.plot(map_widget=pdx_density_map, renderer_type='h')
There is a hotspot near downtown Portland. There are few other isolated hotspots around the downtown as well. These could be pockets of communities that are spread around a major city.
Using different renderers such as 'class colored renderer', you can visualize and investigate if there is any spatial phenomena observable for the following columns.
pdx_age_map = gis.map("Portland, OR")
pdx_age_map.basemap = 'gray-vector'
pdx_age_map

prop_sdf.spatial.plot(map_widget = pdx_age_map,
renderer_type='c', # for classs breaks renderer
method='esriClassifyNaturalBreaks', # classification scheme
class_count=10, # between 1900 - 2000, each decade in a class
col='YEAR BUILT',
cmap='Blues', # matplotlib color map
alpha=0.7,
outline_color=[0,0,0,0])
If you notice the syntax above, it mirrors the general syntax of plotting a DataFrame as a chart. It uses the same color map and symbols of matplotlib. Internally, the ArcGIS API for Python converts this syntax to ArcGIS symbology and renders it on a map.
Since the colormap and symbols is that of matplotlib, you can collect the class breaks from the map above and use that to plot the same data as a bar chart as shown below. Combining the map and chart gives you a powerful way to interactively explore the spatial and statistical distribution of your dataset.
age_class_breaks = pdx_age_map.layers[0].layer.layerDefinition.drawingInfo.renderer.classBreakInfos
# print(len(age_class_breaks))
cbs_list = []
cmap_list = []
for cb in age_class_breaks:
# print(cb.description) # print the class break labels
cbs_list.append(cb.classMaxValue)
cmap_list.append([x/255.0 for x in cb.symbol.color])
# build a histogram for the same class breaks
n, bins, patches = plt.hist(prop_sdf['YEAR BUILT'], bins=cbs_list)
# apply the same color for each class to match the map
idx = 0
for c, p in zip(bins, patches):
plt.setp(p, 'facecolor', cmap_list[idx])
idx+=1
plt.title('Histogram of YEAR BUILT column')
pdx_price_map = gis.map("Portland, OR")
pdx_price_map.basemap = 'gray-vector'
pdx_price_map

prop_sdf.spatial.plot(map_widget = pdx_price_map,
renderer_type='c', # for classs breaks renderer
method='esriClassifyQuantile', # classification scheme
class_count=10, # between 1900 - 2000, each decade in a class
col='PRICE',
cmap='BuPu_r', # matplotlib color map
alpha=0.5,
outline_color=[50,0,0,50], line_width=1)
price_class_breaks = pdx_price_map.layers[0].layer.layerDefinition.drawingInfo.renderer.classBreakInfos
# print(len(age_class_breaks))
cbs_list = []
cmap_list = []
for cb in price_class_breaks:
# print(cb.description) # print the class break labels
cbs_list.append(cb.classMaxValue)
cmap_list.append([x/255.0 for x in cb.symbol.color])
# build a histogram for the same class breaks
n, bins, patches = plt.hist(prop_sdf['PRICE'], bins=cbs_list)
# apply the same color for each class to match the map
idx = 0
for c, p in zip(bins, patches):
plt.setp(p, 'facecolor', cmap_list[idx])
idx+=1
plt.title('Histogram of PRICE column')
pdx_size_map = gis.map("Portland, OR")
pdx_size_map.basemap = 'gray-vector'
pdx_size_map

prop_sdf.spatial.plot(map_widget = pdx_size_map,
renderer_type='c', # for classs breaks renderer
method='esriClassifyNaturalBreaks', # classification scheme
class_count=10, # between 1900 - 2000, each decade in a class
col='SQUARE FEET',
cmap='RdBu', # matplotlib color map
alpha=0.7,
outline_color=[50,0,0,50], line_width=1)
size_class_breaks = pdx_size_map.layers[0].layer.layerDefinition.drawingInfo.renderer.classBreakInfos
# print(len(age_class_breaks))
cbs_list = []
cmap_list = []
for cb in size_class_breaks:
# print(cb.description) # print the class break labels
cbs_list.append(cb.classMaxValue)
cmap_list.append([x/255.0 for x in cb.symbol.color])
# build a histogram for the same class breaks
n, bins, patches = plt.hist(prop_sdf['SQUARE FEET'], bins=cbs_list)
# apply the same color for each class to match the map
idx = 0
for c, p in zip(bins, patches):
plt.setp(p, 'facecolor', cmap_list[idx])
idx+=1
plt.title('Histogram of SQUARE FEET column')
pdx_hoa_map = gis.map("Portland, OR")
pdx_hoa_map.basemap = 'gray-vector'
pdx_hoa_map

#plot properties without HOA as hollow
# prop_sdf_hoa_f = prop_df[prop_df['HOA PER MONTH']==0]
# prop_sdf_hoa_f.spatial.plot(map_widget=pdx_hoa_map, symbol_type='simple',
# symbol_style='+',outline_color='Blues',
# marker_size=7)
prop_sdf_hoa_2 = prop_df[prop_df['HOA PER MONTH']>0]
prop_sdf_hoa_2.spatial.plot(map_widget = pdx_hoa_map,
renderer_type='c', # for classs breaks renderer
method='esriClassifyQuantile', # classification scheme
class_count=10, # between 1900 - 2000, each decade in a class
col='HOA PER MONTH',
cmap='RdBu', # matplotlib color map
alpha=0.7,
outline_color=[0,0,0,0], line_width=0)
hoa_class_breaks = pdx_hoa_map.layers[0].layer.layerDefinition.drawingInfo.renderer.classBreakInfos
# print(len(age_class_breaks))
cbs_list = []
cmap_list = []
for cb in hoa_class_breaks:
# print(cb.description) # print the class break labels
cbs_list.append(cb.classMaxValue)
cmap_list.append([x/255.0 for x in cb.symbol.color])
# build a histogram for the same class breaks
n, bins, patches = plt.hist(prop_sdf['HOA PER MONTH'], bins=cbs_list)
# apply the same color for each class to match the map
idx = 0
for c, p in zip(bins, patches):
plt.setp(p, 'facecolor', cmap_list[idx])
idx+=1
plt.title('Histogram of HOA PER MONTH column')

ax_list = prop_df.hist(bins=25, layout=(4,4), figsize=(15,12))
Explore the frequency of categorical columns
fig2, ax2 = plt.subplots(1,2, figsize=(10,5))
prop_df['CITY'].value_counts().plot(kind='bar', ax=ax2[0],
title='City name frequency')
ax2[0].tick_params(labelrotation=45)
prop_df['PROPERTY TYPE'].value_counts().plot(kind='bar', ax=ax2[1],
title='Property type frequency')
ax2[1].tick_params(labelrotation=45)
plt.tight_layout()
Let us define a few rules based on the intrinsic properties of these houses and shortlist them.
filtered_df = prop_sdf[(prop_df['BEDS']>=2) &
(prop_df['BATHS']>1)&
(prop_df['HOA PER MONTH']<=200) &
(prop_df['YEAR BUILT']>=2000) &
(prop_df['SQUARE FEET'] > 2000) &
(prop_df['PRICE']<=700000)]
filtered_df.shape
(prop_sdf.shape, filtered_df.shape)
From 3624 houses, we shortlisted 331 of them. Below, let us visualize the statistical distribution of this shortlist.
ax_list2 = filtered_df.hist(bins=25, layout=(4,4), figsize=(15,15))
From the histograms above, we notice most of the houses have 4 beds, while we requested for at least 2. Majority of them are newly built and skewed toward upper end of the price spectrum.
pdx_filtered_map = gis.map("Portland, OR")
pdx_filtered_map.basemap = 'gray-vector'
pdx_filtered_map

filtered_df.spatial.plot(map_widget=pdx_filtered_map,
renderer_type='c',
method='esriClassifyNaturalBreaks', # classification scheme
class_count=10,
col='PRICE',
cmap='Blues', # matplotlib color map
alpha=0.7,outline_color=[0,0,0,0])
The houses in the shortlist are well spread across the Portland market. We notice spatial clustering in the distribution of property prices. Higher priced houses are mostly to the west of downtown while moderately and lower priced are spread across east and south.
So far, we used attribute queries to explore and filter out properties. We have not yet used GIS analysis to narrow them further. Before that, let us save our work to disk.
filtered_df.to_csv('resources/houses_for_sale_att_filtered.csv')